The autonomous shuttle was testing today when it was grazed by a delivery truck downtown. The shuttle did what it was supposed to do, in that it’s sensors registered the truck and the shuttle stopped to avoid the accident. Unfortunately the delivery truck did not stop and grazed the front fender of the shuttle. Had the truck had the same sensing equipment that the shuttle has the accident would have been avoided. Testing of the shuttle will continue during the 12-month pilot in the downtown Innovation District. The shuttle will remain out of service for the rest of the day. The driver of the truck was cited by Metro.
[…]
The Dymaxion car was crashed into, too, and similarly unfairly maligned by a headline, ruining its chances.
From Wikipedia:
In the press, no mention was made that the Dymaxion had been involved in a /two/-car accident. Instead, the cause of the accident was attributed to the car’s unconventional configuration: headlines in New York and Chicago read, “Freak car rolls over – killing famous driver – injuring international passengers”.
The subsequent formal investigation found the actual cause of the impact was a collision with a car driven by the Chicago South Park commissioner who wanted a closer look at the Dymaxion — and who immediately left the scene after the accident. According to the official coroner’s inquest, the two vehicles were traveling at 70 mph, with /Turner trying to evade the politician’s car/. The inquest found the design of the Dymaxion was not a factor in the accident.
(But the facts came out too late. The press had made it out to be a killer freak car.)
If I saw someone standing at a corner, looking at their phone, I would treat them as much likelier to suddenly walk out into traffic, and probably slow down considerably.
I know YOU know you aren’t about to walk out into traffic, but I’ve seen too many people do stupid things while staring at their phone to assume someone knows what they’re doing.
yeah, i know. it’s true. but body language can also convey a lot. if i stop and sort of turn with my side of back the intersection, and am doing something with my phone, i would hope drivers get the idea i’m not a rabbit and i’m not going to suddenly dart out into traffic.
As I was just driving around seeing the usual stupidity, I realized why the truck hit the shuttle. The shuttle wasn’t intimidated. Instead of freaking out and saying “that trucks going to hit me!!!” and backing up, it just stopped with the reasonable assumption that the truck wouldn’t actually hit it. But think of how much driving is actually a freaking playground where the bullies are always trying to get their way, and truck drivers are used to throwing their size around to intimidate.
How would the autonomous driver do in the game of serial chicken that is the 9 to 2 lane merge into the Holland Tunnel, where everyone acts like it’s a deathmatch and will stick inches from the bumper of the car in front of them? The computer will freeze and cry “Norman coordinate!!!”
So walking across the street while staring at a smartphone, not even attempting to look at traffic, even if jaywalking, is not “distracted walking?” Because I’ve seen that about a hundred bajillion times, and I’d call it “distracted walking.”
I, too, have driven a lot more than two hours, and have yet to have an equivalent accident. That’s because I have never had a truck back into me. It is not a common thing. Indeed, it is so unlikely, I briefly wondered whether it was an accident at all, and not something to discredit AI. No, it seems it was an accident.
I think a general driving software strategy is to stop in a controlled fashion when something unfamiliar is happening, and to stay stopped until things have become familiar again. This would seem to be a better strategy than anything else. It would certainly be a better strategy than trying to take any evasive action, unless the software has been trained to cover that particular event. What it should not do is make a poor extrapolation from its prior data.
They could have a separate system that would sound the horn if a collision from any direction is predicted.
[ In the interests of full disclosure, I admit I am very tempted to say that the AI system may have no rear facing cameras, or cannot back up; and then go in search of evidence for this. But I am trying to be fair. ]
To be fair, even though i know the story of the asshole who caused that crash, the Dymaxion was fucking terrifying. Have you seen the video of the replica one being driven?
you mean, exactly like when a human driver fails to interpret a pedestrian’s intentions and hurts someone? A thing that happens every day?
and the urban environment where pedestrians are “truly safe” from human drivers is where?
there presently exists some scenario where any given human driver is going to misinterpret one or more human behavior(s) exhibited by one or more pedestrian(s).
The logic you’re using to say AIs aren’t ready means that neither are humans.
You aren’t wrong. The AI does need to get at least as good as the human aggregate of interpreting peds. What I’m pointing out is that the AI is already better at interpreting peds (and all other driving scenarios) than some quantity of human yet legal drivers.
But our argument is insignificant given that human drivers can–and regularly do–lapse to zero for driver ability (drunk, phone, passenger distraction, sleepy, mind wanders, medical issue, glare, blind spot, can’t hear etc). Under these circumstances, there can be no interpretation of a pedestrian’s intentions whatsoever.
